This month’s T-SQL Tuesday is hosted by Tomaz Kastrun – his call is to write about how we’ve used ChatGPT, and what are ethical issues, if any, that we have encountered while using it.
I am a relatively new user. I know most people are like me, but some of us are newer than others at this 🙂 I use the free version of 3.5. I read a lot of posts and tried to understand more of how it works, but I have learned that nothing beats actual experience using it. I learned that it is really good with narrative, and one podcast I listened to recommended using it to train in doing interviews. I haven’t interviewed for a job in 6 years now. I am still not interviewing since I like where I work, but I know it is a skill that gets rusty without practice. So I decided to try it out this way.
I started by introducing myself and asking it to role play an interview for a Database Engineer role, with me. Below is the response I got.
I was looking for it to ROLE PLAY an interviewer, but it put out an entire interview for me, giving me no chance to respond. What is more, some of those lines seem very similar to *my own lines* from many years ago – I’ve used them somewhere, don’t recall where – in a blog post, or in a podcast interview. It was me, most of that. I found this rather creepy. But I continued to ask it to work with me.
I got really hopeful now. I told it what our roles are and it has clearly told me in return that it understood, so what would the response be?
Ugh. Same thing again. I didn’t get a chance to say a word, not a word. It did the entire interview by itself, this sorta thing doesn’t help much at all.
Again, no luck. Clearly reprimands don’t work very well. So I decided to give it one final shot, and it worked!
So there it was! From this point on, it was fun. Not all questions or answers were ‘correct’ or up to snuff technically, but that wasn’t the point of the exercise. The point was to get some basic practice like you would talking to a person. To get it to do this – the issue was finding the right word to grab and understand what I need. Word, not sentences, like humans would. The word, in this case, was ‘question’. After I asked to ask me a question, it was on a roll, going on and on with questions.
I think this was too basic a test to find any serious issues, but I had one concern.
In the initial round, when it spit out full interviews, a lot of verbiage was mine, but I am myself not sure where it found it. No attribution was given. I am also not sure if this verbiage would appear for someone else, like is it reusing something I said for someone else? Not a big deal, it is only an interview practice, but still, would be nice to know. More importantly, how did it know it was me? Via my email?
I asked about where it found the technical content , and the answer was what one would expect.
To conclude, I found an amazing quote in a book I read recently, for school – ‘A word is known by the company it keeps’. I think this is a phrase that is very applicable to ChatGPT. Find the right words to use with it, and limit your expectations on what you get – I don’t expect a high degree of SQL Server expertise, I did expect a conversational interview with basic answers, which can be helpful to practice, and it lived up to that.
Thank you Tomaz, for hosting.